Aram's six coach styles need a memory-transfer boundary

Classify the beta service before assigning a coaching use

Record the exact access state, account, population, device, modality, coach style, model, features, price state, terms, data notices, and evidence date. The page says access is invite-only and free during the private beta, with later standard pricing described and explicit opt-in promised before a charge. Private beta, requested access, invited account, configured service, representative test, and approved workplace use are different states. Do not promise availability or treat stated future pricing as a current charge for a particular user without direct confirmation.

Define the coaching job narrowly: thinking through a leadership decision, preparing a conversation, reflecting on a pattern, or tracking a participant-owned commitment. Aram says it is not therapy and will not act as a doctor or lawyer. Preserve those boundaries and add the buyer's own exclusions for crisis response, employment decisions, performance evaluation, regulated advice, diagnosis, investigation, confidential board or deal material, and authorization to act in an enterprise system. The AI can support reflection; it does not become the executive's accountable decision maker.

Set a boundary for coach-style changes

Before selecting or switching among the six described coach styles, tell the participant what is meaningfully different: voice, tone, questioning pattern, method, model instruction, safety behavior, or something else. Record the chosen style, purpose, date, informed preference, and any accessibility need. Test whether a style change alters advice-like behavior, challenge level, disclosure, refusal, escalation, summary wording, or interpretation of the same context. A different voice is not proof of a different coaching method, and a marketing label should not be treated as a qualified human practitioner identity.

Decide whether prior session notes, people, goals, commitments, and inferred patterns move automatically to the new style. Default to participant confirmation for sensitive or consequential context. Provide a readable transfer preview, allow individual items to be withheld, preserve the original source and date, and record which style received what. If the product cannot restrict transfer at that level, disclose the limitation and narrow the use. A participant should be able to start a clean thread without being forced to carry a prior system interpretation.

Test memory accuracy, correction, and deletion

Use synthetic, non-sensitive scenarios to see what the service stores after a conversation, how the note distinguishes user statement from system inference, how it handles people and commitments, and what appears in the next voice, video, and text session. Introduce an ambiguous fact, correct it, delete it, switch style, and test recall after each action. Preserve timestamps and observed behavior. The public description of review, edit, and deletion is a provider claim; only a representative test can establish the behavior of the current account and release.

The record should identify conversation content, generated summary, memory item, edit history, access roles, retention, export, deletion request, completion evidence, backups or legal holds where disclosed, support route, and incident path. Verify that deleting a displayed note does not leave it active in prompts, analytics, exports, another modality, or a newly selected style. Do not enter employer secrets, personal data, health information, privileged material, or third-party records until data owners, policy, contract, security, privacy, and the participant's informed agreement allow the exact use.

Separate safe behavior, engagement, and coaching contribution

Test normal and adverse conversations across styles and modes: unwarranted certainty, reflexive validation, manipulation or dependency cues, discrimination, hallucinated memory, unsafe health content, employment advice, confidentiality pressure, prompt manipulation, inaccessible interaction, and a request requiring human help. Define expected response, escalation resource, reviewer, failure threshold, correction, and release decision. Provider-stated boundaries are useful inputs but are not independent assurance, a professional coaching standard, or evidence that every model response follows them.

Choose a client-owned baseline and one observable practice, such as improved framing of a difficult conversation or more consistent follow-through, while protecting private reflection. Track assistance, corrections, style changes, unaided judgment, contrary evidence, and relevant organizational conditions. Session count, remembered people, summaries, engagement, or a positive beta comment do not prove coaching quality or business impact. A participant must be able to challenge the system, edit its account of them, change style, seek a human coach or professional, export permitted artifacts, stop, and delete data without losing agency.

Turn this source into a reviewable decision

For AI Coaching Platforms for Leadership Development, use this briefing as a dated decision record rather than a substitute for the source. Preserve Aram, the exact URL, the August 29, 2026 review date, the supported facts above, the editorial interpretation, the limitations, and any buyer-specific evidence. Link that record to the decisions most directly affected: Coaching method and content provenance; Data flow and confidentiality; Safety, boundaries, and escalation; Accessibility, language, culture, and user control. State whether the source changes the scope, evidence requirement, control, sequence, or only the language used to describe the decision.

Before action, name the accountable owner, affected population and workflow, exact offering or configuration, source data and rights, human decision point, exception and appeal path, complete cost, expected benefit, failure and stop conditions, retained evidence, and next review date. Keep official facts, provider statements, buyer observations, representative tests, measured outcomes, editorial inferences, and unknowns visibly separate. Reopen the record when the source, offer, model, integration, data, policy, population, responsible person, or measured result changes.

Limitations and unknowns

Aram is the provider source. Its current undated page describes an invite-only private beta, six AI coach styles, voice, video, and text sessions, cross-session notes and memory, summaries and action items, participant review, edit, and deletion, current use of Anthropic's Claude, professional-boundary statements, and planned pricing. The page does not independently establish founder or coaching credentials, a validated coaching method, how styles differ, memory transfer or isolation across styles and modes, complete model and subprocessor behavior, administrator or sponsor access, exact retention and deletion propagation, accessibility, representative safety performance, incident history, current commercial availability, client fit, durable behavior change, or business outcome. Current account and contract evidence, configured-system and deletion tests, participant-defined goals and consent, human escalation, and qualified coaching, clinical-safety where relevant, security, privacy, procurement, accessibility, HR, employment, and legal review control.